Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Ecological Indicators 18 (2012) 693–704

Contents lists available at SciVerse ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Relationship of land surface and air temperatures and its implications for
quantifying urban heat island indicators—An application for the city of Leipzig
(Germany)
Nina Schwarz a,∗ , Uwe Schlink b , Ulrich Franck b , Katrin Großmann c
a
UFZ - Helmholtz Centre for Environmental Research, Department Computational Landscape Ecology, Permoserstrasse 15, 04318 Leipzig, Germany
b
UFZ - Helmholtz Centre for Environmental Research, Core Facility “STUDIES”, Permoserstrasse 15, 04318 Leipzig, Germany
c
UFZ - Helmholtz Centre for Environmental Research, Department Urban and Environmental Sociology, Permoserstrasse 15, 04318 Leipzig, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Urban heat islands (UHIs) describe the phenomenon of altered temperatures that occur in urban areas
Received 27 September 2011 when compared to their rural surroundings. UHIs influence human well-being, human health and the city
Received in revised form 2 January 2012 as an ecological niche. UHIs can be quantified with meteorological ground measurements of air temper-
Accepted 4 January 2012
atures or with remotely sensed land surface temperatures (surface urban heat island). Both approaches
have advantages and disadvantages and are rarely combined. Further, within these approaches, different
Keywords:
indicators for quantifying the UHIs are used. In this methodological study, we (1) combined data on land
Urban heat island
surface and air temperatures, (2) enriched the debate by suggesting the application of indicators for the
Indicator comparison
Remote sensing
two distinct data sets and (3) systematically quantified indicators of all approaches for the city of Leipzig,
Ground temperature Germany. A relationship between the land surface and air temperatures was established. However, the
Meteorological study results for the single indicators showed that the absolute values of the detected UHI in Leipzig depend
Data integration on the selected indicator and the data set used. The main conclusion for future studies on UHIs is to use
several UHI indicators in parallel to acknowledge the uncertainty of measuring the UHI using a single
indicator and either ground measurements or remote sensing.
© 2012 Elsevier Ltd. All rights reserved.

1. Introduction and they influence vegetation dynamics in the city, e.g., earlier
bud-bursts (Luo et al., 2007; Roetzer et al., 2000; Shustack et al.,
The term urban heat island (UHI) describes the phenomenon 2009). Increasing mean temperatures that occur due to climate
of altered temperatures in urban areas compared to their rural change add further weight to the importance of the UHI.
surroundings (Oke, 1982). UHIs are defined as the temperature To quantify UHIs and understand the different approaches that
difference in urban and rural areas with diverse indicators for are used in indicator development, it is essential to distinguish
quantifying the difference. Temperature differences result from between UHIs that are quantified (a) from air temperatures and (b)
the influence of the thermal emissivity properties of urban sur- from land surface temperatures (LSTs). (a) Air temperature mea-
faces and the three-dimensional configuration and heat capacity of surements of UHIs refer to the so-called canopy layer, which is the
erected structures onto temperature patterns in an urban region. lowest part of the atmosphere from the ground up to the tree or
Most of the world’s cities thus show higher temperatures in the building height (Arnfield, 2003; Stewart, 2011). In the canopy layer,
urban core than in the surrounding rural areas. One exception is spatially diverse air temperatures are present in urban and rural
arid or semi-arid regions, in which the city cores may be cooler areas. The next layer, which is the boundary layer, is also affected by
due to factors such as irrigated gardens. UHIs directly (Harlan et al., the urban surface but is not usually measured with meteorological
2006; Lafortezza et al., 2009) and indirectly (Stafoggia et al., 2008) data of ground measurements (Oke, 1982). (b) Surface temperature
affect the thermal comfort and health of the cities’ inhabitants. Fur- data for UHI quantifications originate from airborne or satellite-
thermore, altered temperature patterns also influence the urban borne sensors (Voogt and Oke, 2003). These sensors measure the
ecology by providing an altered species habitat (Knapp et al., 2008), temperature of the land surface as seen from above, and this tem-
perature is related to the air temperature in the same location.
Voogt and Oke (2003) therefore proposed the term surface urban
∗ Corresponding author. Tel.: +49 341 235 1970; fax: +49 341 235 1939.
heat island (SUHI) for a UHI that is measured with LST.
Different indicators quantify UHI and SUHI. For instance, air
E-mail addresses: nina.schwarz@ufz.de (N. Schwarz), uwe.schlink@ufz.de
(U. Schlink), ulrich.franck@ufz.de (U. Franck), katrin.grossmann@ufz.de temperatures are used to calculate temperature differences of
(K. Großmann). urban and rural weather stations or temperature traverses through

1470-160X/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ecolind.2012.01.001
694 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

Table 1
Typology of the UHI indicators for LSTs and air temperatures, UHI-driven and land-cover-driven.

Approach Indicator SUHI UHI


Q: quantification with LST; [unit] Q: quantification with AIRT; [unit]
S: source S: source
CS: adaption to case study CS: adaption to case study

Magnitude Q: (maximum LST of all pixels) – (mean LST of all Q: (maximum AIRT of all measurements) – (mean AIRT
pixels); [K] of all measurements); [K]
UHI-driven
S: Rajasekar and Weng (2009) S: new suggestion
CS: – CS: measurements from all weather stations
Range Q: LST [maximum LST of all pixels] – LST [minimum Q: AIRT [maximum AIRT of all measurements] – AIRT
LST of all pixels]; [K] [minimum AIRT of all measurements]; [K]
S: new suggestion S: Garcia-Cueto et al. (2007) and Runnalls and Oke
CS: – (2000)
CS: measurements used (1) from all weather stations,
(2) from mobile route
Gradient Q: unstandardised linear regression coefficient for all Q: unstandardised linear regression coefficient for
LST ← distance from city centre; [K/km] mobile AIRT ← distance from city centre; [K/km]
S: new suggestion S: new suggestion
CS: city centre = central station CS: city centre = central station
Hot island area Q: total area with [LST > (mean LST + 1 standard Not applicable
deviation)]; [km2 ]
S: Zhang and Wang (2008)
CS: –
Urban-rural (density) Q: LST [pixel with high density of sealed surface] – LST Q: AIRT [weather station in high density area] – AIRT
[pixel with low density of sealed surface]; [K] [weather station in low density area]; [K]
Land-cover-driven S: Gallo et al. (1993) S: Gallo et al. (1993) and Kukla et al. (1986)
CS: LST [pixel on station no. 1] – LST [pixel on station CS: AIRT [station no. 1] – AIRT [station no. 15]
no. 15]
Urban-rural (distance) Q: LST [pixel at the city centre] – LST [pixel at the Q: AIRT [weather station at the city centre] – AIRT
periphery]; [K] [weather station at the periphery]
S: Tomlinson et al. (2010) S: new suggestion
CS: LST [pixel on station no. 8] – LST [pixel on station CS: AIRT [station no. 8] – AIRT [station no. 14]
no. 14]
Urban-agriculture Q: mean LST [all urban covers] – mean LST Q: AIRT [weather station in urban area] – AIRT
[agriculture]; [K] [weather station in agricultural area]
S: Jin et al. (2005) S: new suggestion
CS: urban classes: urban atlas codes 11100–11240. CS: AIRT [station no. 1] – AIRT [station no. 14]
Agricultural class: urban atlas code 20000
Urban-water Q: mean LST [all urban classes] – mean LST [water]; [K] Not applicable
S: Chen et al. (2006)
CS: urban classes: urban atlas codes 11100–11240.
Water class: urban atlas code 50000
Micro-urban heat island Q: percentage of area without water surfaces with Not applicable
[LST > maximum LST of forest classes]; [%]
S: Aniello et al. (1995)
CS: forest class: urban atlas code 30000. Water class:
urban atlas code 50000

AIRT, air temperature; LST, land surface temperature. Not applicable: indicator cannot be transferred from AIRT to LST or vice versa. CS: - no adaptation to the case study
was necessary to quantify the indicator.

a city (Stewart, 2011). LSTs are used to calculate diverse SUHI the applicability of UHI/SUHI indicators for air temperatures and
indicators, such as temperature differences for urban and agricul- LSTs and (2) compared UHI/SUHI indicators that stem from these
tural pixels or indicators that attempt to capture the spatial pattern approaches. Thus, we hypothesize that
of temperatures (Schwarz et al., 2011). Both approaches have spe-
cific advantages. Air temperatures are a direct UHI measure, but • Air temperatures and LSTs are related (hypothesis 1).
they are usually only available for single measurement stations or • Indicators to quantify the UHI/SUHI deliver comparable results
traverses through a city. Conversely, SUHI can account for the tem- (hypothesis 2).
perature distribution within a large area, but it is only an indirect
estimate of the UHI. Therefore, the combination of both approaches In this study, these indicators are differentiated regarding the
provides added value for the description of UHIs. data that were used (which led to UHI and SUHI indicators, respec-
A variety of studies have combined LST and air temperature data tively). Furthermore, indicators are distinguished regarding the
(e.g., Garcia-Cueto et al., 2007; Hartz et al., 2006; Lindberg, 2007; differentiation of what is considered “urban” and “rural” (Stewart
Voogt and Oke, 2003). However, specific indicators are solely cal- and Oke, 2009). We distinguish between the UHI-driven approach
culated with either air temperatures or LSTs. The only exception and the land-cover-driven approach (Table 1). The UHI-driven
seems to be the study by Gallo et al. (1993), who compared the approach uses temperature patterns to distinguish urban and rural
UHI/SUHI urban-rural temperature difference. No study was found areas (Lowry, 1977); the rural areas are those areas that are not
that compared both different indicators for UHI/SUHI with air and influenced by the UHI and therefore show lower temperatures,
LST in parallel. Our study aims at filling this gap by comparing indi- while the urban areas show higher temperatures because of the
cators for UHI/SUHI with a combination of air temperature and LST UHI. The land-cover-driven approach a priori defines urban and
data. This methodological study shall provide insights into the com- rural areas by means of land cover data that differentiate between
parableness of indicators and data sets that are frequently used for urban (primarily built-up surfaces and their surroundings) and
UHI/SUHI descriptions. Consequently, the present study (1) tested rural (primarily natural or agricultural) areas and determines the
N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704 695

temperature differences in the second step. For both approaches, sea level. The landscape around the city is dominated by agriculture
UHI and SUHI indicators are available. and former opencast lignite mining. Today, most of these mines are
A UHI-driven indicator that is based on air temperatures is the being re-cultivated and converted to lakes, especially in the south
range (equal to the difference between the maximum and the min- of Leipzig. The Weisse Elster River and contributing streams flow
imum) of temperatures that are compiled in mobile measurements through the city and are partly accompanied by a Riparian forest
or with weather stations (e.g., Garcia-Cueto et al., 2007; Runnalls that intersects the region in north–south direction. The climate in
and Oke, 2000). Furthermore, a linear regression of air tempera- the case study region is continental with an average annual tem-
tures of mobile measurements shows that the cooling rate depends perature of 9 ◦ C and contains clearly distinguishable seasons; the
on the distance to the city centre (an indicator suggested here for average winter (DJF) temperatures were 0.5 ◦ C and the summer
the first time). The corresponding SUHI indicators are the hot island (JJA) temperatures were 17.4 ◦ C for the years from 1961 to 1990.
area (equal to the total area with an LST higher than the mean plus The mean annual precipitation is approximately 540 mm.
one standard deviation; Zhang and Wang, 2008) and the magnitude The data set consisted of air temperature measurements and
(equal to the maximum of all temperature pixels minus the mean LST maps. The LSTs were detected with an airborne sensor dur-
of all temperature pixels; Rajasekar and Weng, 2009) because they ing two flights (Section 2.3). The air temperatures were acquired
both solely rely on temperature patterns and do not account for from a total of 15 weather stations from three different origins
additional land cover data when quantifying the SUHI. (Section 2.2.1) and mobile measurements parallel to the evening
The land-cover driven indicator that is based on air tempera- flight (Section 2.2.2). Two flights were utilised: (1) shortly after
tures is the difference of the temperature at the urban and rural sunset on 22 September 2010 between 7:30 pm and 9:00 pm (called
weather stations because these pairs of stations are often selected “evening” for the remainder of this paper) and (2) before sunrise on
according to population density to distinguish urban and rural areas 23 September 2010 between 5:00 am and 6:30 am (called “morn-
(Gallo et al., 1993; Kukla et al., 1986). The corresponding SUHI indi- ing” throughout the paper). All of the times are given in Central
cator identifies rural and urban weather stations using land cover European Summertime (UTC +2 h). During these days, in the case
or the distance to the centre and uses the LSTs of the correspond- study region, there was autochthone high-pressure weather with
ing pixels (Gallo et al., 1993; Tomlinson et al., 2010 for the rural a clear sky. The winds were very weak (German Weather Service
area). Furthermore, land cover information is used to compute the observations at Leipzig-Schkeuditz weather station: air pressure:
difference of urban and agricultural land cover (Jin et al., 2005), 1006 hPa, mean air temperature 2 m above the ground 14.7 ◦ C (min:
the difference of urban and water pixels (Chen et al., 2006) and the 8.0 ◦ C, max: 21.4 ◦ C), relative air humidity 78.1%, average wind 2 Bft,
micro-urban heat island (equal to the percentage of area (without wind gusts 4 Bft, sunshine duration 11.2 h, no precipitation).
water surfaces) with an LST higher than the warmest LST that is
associated with tree canopies; Aniello et al., 1995). 2.2. Air temperature data
These diverse indicators were developed and used in two mostly
separated research communities (with Gallo et al., 1993 being a rare 2.2.1. Stationary measurements
exception); these two communities are largely characterised by the A total of 15 weather stations were used to acquire air temper-
data that were used (either remote sensing data or meteorologi- atures (Figs. 1 and 2); two stations (nos. 14 and 15 in Fig. 2) are
cal data). The indicators they use to describe the UHI/SUHI depend operated by the German Weather Service; four stations (nos. 4, 8,
on the type of data that are gathered, e.g., some LST-based indica- 12, and 13 in Fig. 2) by the Saxon State Office for Environment,
tors require spatially explicit data that cover the whole area under Agriculture and Geology; and nine stations were mounted specif-
investigation. However, the indicators that have been developed in ically for this study. These nine stations (1, 2, 3, 5, 6, 7, 9, 10 and
these two communities may potentially affect each other because 11) were equipped with a temperature and humidity sensor and a
the ideas underlying a specific indicator can be transferred. For data logger (OPUS10 TIC, Lufft company, Germany, accuracy ±0.3 K
instance, in the literature, it was proposed that the UHI magnitude and ±2.5% relative humidity) and were mounted at a 1.5 m height
should be used with LST data, but the UHI can also be quantified in a ventilated shelter that protected against solar radiation and
using maximum and mean air temperatures. Consequently, when- precipitation.
ever possible, we attempted to adapt the different indicators to
both air temperatures and LSTs. Table 1 shows the typology of the 2.2.2. Mobile measurements
UHI indicators that are either driven by the UHI or by land cover The mobile measurements were performed on a route that is
and are quantified using LSTs or air temperatures. west of the city centre and shows a gradient from dense urban
This approach was realised for the city of Leipzig, Germany. The structures into a large green urban area (Fig. 3). The route was
focus was on UHI/SUHI indicators that describe the UHI in single furthermore designed to include roads with different tree covers
metrics to foster comparisons across cities or studies. Thus, the (Fig. 3; starting from the east, the three parallel streets have numer-
approach of using geographically weighted regression techniques ous trees in a wide street, numerous trees in a narrower street and
for single cities (e.g., Li et al., 2010) was not pursued further because no trees). The route from the city centre to the park was approxi-
it provides a very detailed view on a single city without quantifying mately 4.5 km one-way and was measured once in each direction,
the absolute extent of the SUHI. starting from the city centre at 7:00 pm (a total time of approx-
imately 2 h). Instruments were located in a backpack that was
carried by one person. Sensors were placed outside the backpack
2. Materials and methods at a height of approximately 0.9 m above the ground. The routes
were recorded by a GPS device (Garmin). The air temperature and
2.1. Case study humidity were measured using a Q-Trak 8552 monitor (company
TSI Inc.) with an accuracy of ±0.6 K for air temperature and ±3.0%
The case study region covers an area of 747 km2 and is situated for relative humidity.
in Central Germany (Fig. 1). It encompasses the administrative area The air temperatures were corrected for decreasing tempera-
of the city of Leipzig (298 km2 ), which contained approximately tures due to progressing time so that they would not confound
519,000 inhabitants in 2009. Thus, it is one of the major cities air temperature differences due to changing surroundings with
of Germany. The case study region lies in the Eastern part of the temperature differences because of evening cooling. This was
North German Plain. It is flat and lies approximately 118 m above accomplished by estimating a linear regression for the stationary
696 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

Fig. 1. The case study region Leipzig. Land cover refers to the urban audit data and is available for the larger part of the region (Section 2.4). The numbers indicate the location
of the weather stations (Fig. 2).

measurements of air temperature at the closest weather station AADS 1250, which uses wavelengths from 8 to 14 ␮m. The resulting
(no. 3) at the time of the mobile measurement. A cooling of 0.016 K LST map has a spatial resolution of 5 m with an extent of 30.25 km
per minute was found (corrected R2 = 0.95). east–west and 24.7 km north–south (Fig. 4). The LSTs were classi-
fied into classes that had a width of 1 ◦ C.
2.3. LST data
2.4. Land cover data
The LSTs were acquired during the two overhead flights on 22
(7:30–9:00 pm) and 23 (5:00–6:30 am) September 2010 at 2000 m To characterise the surroundings of the stationary and mobile
above the ground. The sensor to detect the LSTs was a Daedalus measurements and the whole case study region, high-resolution
N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704 697

Fig. 2. Weather stations used for this study: ID and name, geographical coordinates and elevation. The stations are arranged in a descending order according to the share of
sealed and continuous urban fabric. All of the coordinates are given in the Gauss-Krueger coordinate system, zone 4. Land cover 2006 refers to the urban atlas classes (see
Fig. 1), and the circles represent buffers of 5, 10 and 100 m around the weather stations. Elevation was determined using the GTOPO dataset from the US Geological Survey.

urban land cover data available from the Urban Atlas project of the onto air temperatures of the mobile route and the corresponding
European Environment Agency were used (EEA, 2010). The vector LSTs (dependent variables).
data are based upon data that have a spatial resolution of 2.5 m and The diverse UHI/SUHI indicators provide single values for the
a minimal mapping unit of 0.25 ha for the urban classes. The current UHI or SUHI of the city of Leipzig, respectively. Thus, only one
version depicts the year 2006 ±1 year and is available for more than data point was available for each specific indicator per acquisi-
200 European city regions. Urban Atlas land cover distinguishes tion time (morning or evening). Consequently, the comparableness
20 classes, of which 17 are urban (see Fig. 1). The accuracy was of the indicators (hypothesis 2) was shown by comparing their
quantified as 87.5% for the combined classes for the Leipzig data absolute values, and no statistical method was used for comparing
set. them.

2.5. Statistical methods


3. Results and discussion
The relationship of LST and air temperatures (hypothesis 1) was
measured with Spearman rank correlations, as they also account The results are first reported and discussed regarding the rela-
for non-linear relationships. Spearman rank correlations were com- tionship of the LST and air temperature (hypothesis 1, Section 3.1).
puted for whole case study area as well as separated by land cover In Section 3.2, the quantification of the different UHI/SUHI indica-
class. Furthermore, an analysis of variance (ANOVA) was computed tors (hypothesis 2) is presented. Finally, we discuss the limitations
to quantify the effect of land cover class (independent variable) of our study (Section 3.3).
698 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

Fig. 3. Map of the air temperatures that were measured during the evening flight with underlying urban land cover (see Fig. 1). The data are already corrected for cooling
during the time of acquisition (see Section 2.2.2). The grey area of the inserted map shows the detailed area.
N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704 699

Fig. 4. LSTs that were acquired on the evening of 22 September 2010.

3.1. Data integration of air and LST measurements pixel of the weather station generally indicated a closer relationship
to the respective air temperature at the weather station.
First, similarities between the air temperatures and LSTs are To quantify the similarity of the air and LST, Spearman rank
tested (Section 3.1.1). Second, the land cover classes are charac- correlations were computed using the LST on the pixels of the
terised with the air temperatures and LSTs (Section 3.1.2). temperature measurements’ locations along the mobile route. The
correlations for air temperature and LST were all statistically sig-
nificant (p < 0.01) and depended on the quantification of the LST;
3.1.1. Integration of the measurement methods the correlation decreased with an increasing area that was cov-
When comparing the UHI and SUHI indicators, the different ered to compute the mean LST (individual pixel: rs = 0.63; buffer of
measurement methods and their implications must be considered. 5 m: rs = 0.62; buffer of 10 m: rs = 0.60; buffer of 100 m: rs = 0.55).
Air temperatures are measured at single points (here, at weather Finally, we also included a moving average LST on a pixel basis by
stations) or at points along a line (here, the mobile route), while LSTs combining the immediate pixels of the measurement location ±
cover a larger area when they are measured with an airborne sen- four locations of the mobile route and found a much higher corre-
sor. Thus, these different measurement types need to be compared. lation of rs = 0.85. This finding is easily explained by the area that is
Fig. 5 shows the air temperatures that were measured at all of the being taken into account compared to the individual pixel; concen-
weather stations parallel to both flights and a set of LSTs: (1) the LST tric buffers around the pixel include portions of the street canyon,
on the pixel at which the weather station was located and the mean rooftops and possibly backyards and neighbouring streets, and this
LST values for buffers of (2) 5 m, (3) 10 m and (4) 100 m around the depends on the size of the buffer. Thus, increasing the buffer size
weather station. The results from the evening flight indicate that increases the complexity of land cover; consequently, correlations
the air temperatures were higher than the LSTs for almost every between the point measurements of the air temperatures and LSTs
weather station with the exception of no. 7, which was situated on decrease. However, the correlation between air temperature and
a lawn. During the morning flight, almost no differences between LST was highest for the moving average LST. The moving average is
the air temperature and LST were visible. The only exception was the mean of the neighbouring pixels of air temperature measure-
weather station no. 9, where creeping plants covered large parts ments within the street canyon. It combines the effect of similar
of the courtyard in which the weather station was located. Con- land cover with advection; for example, temperatures in a street
sequently, the morning temperatures of the air and land surfaces canyon are also influenced by wind and are likely to be more similar
were closely related, and the LST that was measured for the specific when exposed to the same wind conditions.
700 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

Fig. 5. Relationship between the stationary air temperature measurements and the related LSTs of concentric rings around the stations, which are shown as a range between
the evening (upper end of the bar) and morning (lower end of the bar) temperatures. The air temperatures are given for 8 pm and 6 am, respectively.

Thus, hypothesis 1 (“Air temperatures and LSTs are related”) 3.1.2. Effects of land cover
has been shown to be valid in this case study. These findings are The correlations between air temperatures and LSTs for the
consistent with a large body of literature that has documented mobile route were altered when differentiating for land cover
relationships between air temperatures and LST values in general classes; for pixels with continuous urban fabric, the correlation
(Mostovoy et al., 2006; Prihodko and Goward, 1997) and for urban between the air temperature and the moving average LST was sim-
areas (e.g., Cheng et al., 2008). Conversely, Hartz et al. (2006) found ilar (rs = 0.64; p < 0.01; N = 60). The correlation was lower for green
similar characteristics of neighbourhoods in Phoenix, Arizona only urban areas (rs = 0.54; p < 0.01; N = 28), while the correlation was
at night, whereas advection in this desert environment likely led to higher than average for roads and their associated land (rs = 0.81;
large differences between the daytime air temperatures and LSTs. p < 0.01; N = 31).
N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704 701

Table 2
LSTs that were differentiated by land cover class. Values were computed using the whole case study area.

ID of land cover Name % in case study Evening: LST mean (SD) Morning: LST mean (SD) Diff. mean
type

11100 Continuous urban fabric (S.L. > 80%) 2 15.2 (1) 10.6 (0.9) 4.6**
11210 Discontinuous dense urban fabric (S.L.: 50–80%) 8 14.7 (1.1) 10.1 (0.8) 4.6**
11220 Discontinuous medium density urban fabric (S.L.: 30–50%) 2 14.3 (1) 9.9 (0.9) 4.4**
11230 Discontinuous low density urban fabric (S.L.: 10–30%) <1 14.2 (1.1) 9.9 (0.9) 4.4**
11240 Discontinuous Very low density urban fabric (S.L. < 10%) <1 14.4 (1) 10.2 (0.6) 4.2**
11300 Isolated structures <1 14 (1) 9.7 (1) 4.3**
12100 Industrial, commercial, public, military and private units 8 14.7 (1.3) 10 (1.1) 4.7**
12210 Fast Transit roads and associated land 1 15 (1.1) 10.3 (0.7) 4.7**
12220 Other roads and associated land 3 15 (1.1) 10.3 (0.9) 4.7**
12230 Railways and associated land 1 15 (1) 9.9 (0.8) 5.1**
12400 Airports 1 14.3 (1.3) 10.7 (0.5) 3.6**
13100 Mineral extraction and dump sites 3 14.5 (1.2) 10.9 (2.2) 3.6**
13300 Construction sites 1 14.5 (1.3) 10.3 (0.9) 4.2**
13400 Land without current use <1 14.4 (1.2) 9.9 (1.1) 4.5**
14100 Green urban areas 4 14.9 (1.1) 10.5 (1.1) 4.4**
14200 Sports and leisure facilities 4 14 (1) 9.7 (1) 4.4**
20000 Agricultural, semi-natural and wetland areas 54 13.6 (0.6) 9.2 (1) 4.3**
30000 Forest 5 15.2 (0.9) 10.6 (0.9) 4.6**
50000 Water 3 16 (1) 14.6 (1.8) 1.4**

Notes: Code no. 12300 (Port Areas) does not exist in the case study area.
**
Difference between the morning and evening LSTs differs significantly from 0 (p < 0.001). To compute these tests, 200 random points were distributed per land cover
class to reduce the effects of spatial correlation in a pixel-by-pixel comparison.

The land-cover classes clearly showed different temperatures results of the LSTs for the whole case study area are summarised
(Fig. 6). The corrected air temperatures for 8 pm were always higher in Table 2. The LST values for the land cover classes are similar to
than the LSTs. However, both temperatures showed the same pat- those that have been published in the literature (e.g., Chen et al.,
tern of relatively cooler or warmer areas throughout the mobile 2006; Gluch et al., 2006; Kottmeier et al., 2007; Weng et al., 2007).
traverse. Even small green areas (e.g., at approximately 2.500 and An analysis of variance (ANOVA) for air temperatures of the
6.500 m) showed strongly decreasing temperatures. Furthermore, mobile route and the corresponding LSTs was performed to test
the specific composition of the green areas was relevant; the return the influence of land cover classes as an independent factor on
of the mobile measurement (after approximately 4.800 m) began temperature. The land cover classes showed statistically significant
after crossing a meadow and walking under trees at the edge of (p < 0.01) influences for both temperatures. The post hoc tests indi-
a wooded area. Thus, the land cover class “green urban areas” cated the same patterns for air temperatures and LSTs (Fig. 7). In
also showed complex temperature patterns. Finally, the descriptive terms of the temperatures, discontinuous urban fabric and green
urban areas both differed significantly from each other and from
all of the other land cover classes.

Fig. 6. Gradient of air temperatures and LSTs for mobile measurements in the
evening. Note: the land classes that were covered with the mobile measurements
varied slightly before and after the return because even small changes (e.g., using
the opposite pavement) may result in another pixel that was covered due to the fine Fig. 7. Significant differences in the means of the corrected air temperatures (lower
resolution (2.5 m) of the land-cover data set. The moving average of the LST consists left) and LSTs (upper right) by land cover class as measured by the mobile measure-
of the individual pixel ± 4 pixels. ments. Grey cells indicate significant differences (p < 0.001).
702 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

Fig. 8. Values for the UHI/SUHI in Leipzig according to the different indicators, data and times of the day.

3.2. Comparison of the UHI/SUHI indicators the temperature decrease was overestimated for the mobile
measurement.
The results for the individual indicators (Section 3.2.1) are fol- • The hot island area according to the LSTs was larger in the evening
lowed by elaborations on the choice of weather stations or pixels (224 km2 ) than in the morning (84 km2 ).
(Section 3.2.2). Finally, we discuss the robustness of the indicators • The difference between the urban and rural temperatures accord-
(Section 3.2.3). ing to the density of the built-up area depicted a congruent
diurnal pattern for air and LST; the difference for the morning
UHI and SUHI was smaller.
3.2.1. Individual indicators
• The difference between the urban and rural temperatures accord-
The different UHI and SUHI indicators showed very different
ing to the distance to the city centre was stable (2.9–3.1 K) with
absolute values for the UHI extent in the case study (Fig. 8). These
the exception of the morning SUHI (1 K).
values were clearly dependent on the indicator itself, the data, and
• The difference between the urban and agriculture temperatures
the time of day.
showed a small range across the measurements and points in
time (1–1.5 K) with the exception of the evening air temperature
• The indicator magnitude showed a higher SUHI in the morning measurements (0.2 K).
than in the evening, while the diurnal trend was opposite for the • The difference between the urban and water temperatures
UHI that was quantified with air temperatures. according to the LSTs showed a cooling effect of −1 K and −4 K
• The indicator range delivered the highest values of all of the indi- for the evening and morning, respectively.
cators that quantify the SUHI/UHI in K. For LST, the range was • The micro-urban heat island was larger for the evening than for
higher in the evening than in the morning, while the relationship the morning.
was reversed for the air temperatures.
• Using the mobile measurements, the linear regression for the Consequently, the absolute values of these diverse indicators
distance to the central station that determined the air tem- cannot easily be compared; and hypothesis 2 (“Indicators to quan-
perature was statistically significant (corrected R2 = 46%), and tify the UHI/SUHI deliver comparable results”) needs to be rejected.
the estimated cooling effect was −2 K/km. For the LSTs, the However, a clear trend, which reveals increased temperatures in
linear regression of the distance to the central station that the city, was visible; this was shown by the diverse temperature
influenced the evening LST was also statistically significant (cor- differences and the linear regression. The only exception was the
rected R2 = 88%). However, the estimated cooling effect was difference between the urban and water temperatures, which indi-
only −0.09 K/km. This difference is very likely an effect of the cated the warmer surface temperatures of water surfaces than in
choice of the mobile route; the route started in the city cen- urban areas. This is the result of the LST measurement in autumn
tre, and the turning point was in a green urban area. Therefore, because the water accumulated heat during the summer.
N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704 703

study. This finding is consistent with results from previous studies


on the comparisons of SUHI indicators across cities (Schwarz et al.,
2011) or across time (Streutker, 2002).
The indicators of the difference between urban and rural (den-
sity and distance) are worthy of future study because they offer
rather stable values for both measurement types and for the
evening and morning measurements. This finding is related to the
study by Gallo et al. (1993), who analysed the relationship of tem-
perature differences of (1) minimum air temperatures of urban and
rural weather stations and (2) the respective daytime LSTs on the
corresponding pixels for 37 cities. They found stronger relation-
ships for the urban and rural weather stations for the stations with
an elevation difference of less than 500 m; this assumption was met
in the present study. However, the results of these indicators clearly
depend on the pairs of weather stations that are selected, which
was shown in Section 3.2.2. Thus, future studies should report sev-
eral UHI/SUHI indicators in parallel to clarify the range of possible
values of the UHI in their case study.

3.3. Limitations of our study

Some study limitations should be mentioned. First, the focus of


this study was on one single evening and the following morning
to cover the times of the overhead flight. A time series of over-
head flights combined with air temperature measurements would
Fig. 9. UHI for air temperatures at the 15 weather stations in this sample. The num- be even more helpful to unravel the relationship between LSTs and
bers refer to the IDs of the weather stations (see Fig. 2). The larger the points, the air temperature, the different indicators for the diurnal cycle and
larger the distance of this weather station to the central station (station no. 8). Dis-
across seasons and under different weather conditions. Second, our
tances between stations along the x-axis indicate the respective UHI in the evening,
while distances along the y-axis indicate the respective UHI in the morning. study design included a mobile route with a turn-around point in
a green urban area. This clearly increased the UHI effect that was
found in the air temperature measurements. Designing a mobile
3.2.2. Choice of weather stations or pixels
route along a traverse through the city within the built-up area
Some of the indicators that were based on land cover are directly
could be an alternative that would facilitate the estimation of the
influenced by the choice of pairs of weather stations or pixels. In
effect of distance alone and would not confound the distance-effect
a lot of the studies, only very few weather stations are available
with land cover-effects. Finally, we were not able to confirm the
for a single city; therefore, the question of choosing a meaning-
data quality of the LST data because we had no access to the raw
ful combination is often not raised. In the present study, we used
data that were acquired during the flights. However, we did not
a total of 15 weather stations. Fig. 9 shows that the UHI as the
encounter obvious flaws in the processed LST values.
temperature difference between a pair of weather stations varied
considerably. While the weather stations with very high tempera-
tures were close to the city centre, other weather stations that are 4. Summary and conclusions
also close to the city centre delivered temperatures that were com-
parable to more remote sites. The continuous urban fabric around This study aimed at comparing LST and air temperature data
the weather stations showed ambiguous influences. and their usage for different indicators of the UHI/SUHI. Such
This finding clearly questions the use of only two weather sta- a comparison could deliver insights into the comparableness of
tions for any UHI study. The specific location of a weather station is different UHI/SUHI indicators and the related data sets that are fre-
seldom varied in the literature, although the problem of urban and quently used in urban climate studies. The present methodological
rural weather stations has already been discussed by Lowry (1977). approach was tested in the case study of Leipzig, Germany. Two
Thus, Runnalls and Oke noted that “The greatest potential for errors overhead flights of the city delivered LSTs. The flight campaign
lies in the choice of the rural site” (Runnalls and Oke, 2000, p. 285). was accompanied by field measurements of air temperatures at
Our results showed that the choice of the urban site should also be 15 weather stations and a traverse in the city. These two types of
critically discussed. data sets were used to quantify indicators for the (surface) UHI
in parallel. We used existing indicators from the literature and
3.2.3. Robustness of indicators applied indicators that were developed for either land surface or
The UHI/SUHI indicators clearly showed a variety of absolute air temperatures to the opposite data set, if possible. In doing so,
values for the UHI in the case study. Approximately half of the SUHI we integrated the two mostly distinct research methods on the
indicators reported a larger SUHI in the evening than in the morn- UHI. Our results showed the following. (1) The relationship of the
ing, while the other half detected a larger SUHI in the morning. The air and LSTs was high. In detail, the pixel-by-pixel correlation of
UHI indicators consistently reported a larger UHI in the evening. the measured air temperatures and the LSTs was higher than when
However, the magnitude of the UHI that was measured with air correlating air temperature point values with concentric buffers of
temperatures was also very different for the different indicators. LSTs. However, the highest correlation was found for the air tem-
Although all of the indicators should answer the same question perature point measurements and the LSTs that were averaged over
about the UHI, they clearly provide very different answers. Our find- the neighbouring measurement points of the traverse. (2) The land
ings suggest that the absolute values that are derived from different cover types showed statistical differences for both the land surface
indicators should not be used for comparisons across cities because and the air temperatures. (3) The absolute values for the UHI heavily
they are unable to provide comparable results even in the same case depended on the indicator that was used, the data and the time of
704 N. Schwarz et al. / Ecological Indicators 18 (2012) 693–704

the day. No dominant absolute value was identified for the extent of EEA, 2010. GMES Urban Atlas Data., http://www.eea.europa.eu/data-and-
the UHI in Leipzig. (4) The remote sensing data (LST data) will more maps/data/urban-atlas.
Gallo, K.P., McNab, A.L., Karl, T.R., Brown, J.F., Hood, J.J., Tarpley, J.D., 1993. The use
reliably identify heat islands in areas that have a dense urban fabric of NOAA AVHRR data for assessment of the urban heat-island effect. JApMe 32,
that includes streets than in areas where the population density is 899–908.
lower. Garcia-Cueto, O.R., Jauregui-Ostos, E., Toudert, D., Tejeda-Martinez, A., 2007. Detec-
tion of the urban heat island in Mexicali, B. C., Mexico and its relationship with
The parallel quantification of a variety of indicators for the UHI land use. Atmosfera 20, 111–131.
using the same case study and data sets guaranteed that the exper- Gluch, R., Quattrochi, D.A., Luvall, J.C., 2006. A multi-scale approach to urban thermal
imental and environmental conditions were equal. However, the analysis. Remote Sens. Environ. 104, 123–132.
Harlan, S.L., Brazel, A.J., Prashad, L., Stefanov, W.L., Larsen, L., 2006. Neighborhood
range of the absolute UHI values varied considerably. This finding microclimates and vulnerability to heat stress. Soc. Sci. Med. 63, 2847–2863.
questions the usage of single values for the UHI for a whole city Hartz, D.A., Prashad, L., Hedquist, B.C., Golden, J., Brazel, A.J., 2006. Linking satel-
because the explanatory power of such an indicator (i.e., “For city lite images and hand-held infrared thermography to observed neighborhood
climate conditions. Remote Sens. Environ. 104, 190–200.
X, we quantified the urban heat island as 3.7 K.”) is rather limited.
Jin, M.L., Dickinson, R.E., Zhang, D.L., 2005. The footprint of urban areas on global
However, when comparing different cities, it is necessary to use climate as characterized by MODIS. J. Clim. 18, 1551–1565.
such single values to quantify influences (i.e., the influence of pop- Knapp, S., Kuhn, I., Schweiger, O., Klotz, S., 2008. Challenging urban species diversity:
ulation size or density, the share of green areas or NDVI and the contrasting phylogenetic patterns across plant functional groups in Germany.
Ecol. Lett. 11, 1054–1064.
elevation or latitude) onto the extent of the UHI. Stewart (2011) Kottmeier, C., Biegert, C., Corsmeier, U., 2007. Effects of urban land use on surface
argued that UHI studies that use meteorological data are often temperature in Berlin: case study. J. Urban Plan. Dev. ASCE 133, 128–137.
not comparable because they seldom provide crucial information, Kukla, G., Gavin, J., Karl, T.R., 1986. Urban warming. JCAM 25, 1265–1270.
Lafortezza, R., Carrus, G., Sanesi, G., Davies, C., 2009. Benefits and well-being per-
such as acquisition times or instrumentation. Furthermore, envi- ceived by people visiting green spaces in periods of heat stress. Urban Forest.
ronmental (such as weather) and experimental conditions should Urban Green 8, 97–108.
be considered. We expand this argument by indicating our findings Li, S.C., Zhao, Z.Q., Xie, M.M., Wang, Y.L., 2010. Investigating spatial non-stationary
and scale-dependent relationships between urban surface temperature and
with regard to the different indicators. If one intends to compare environmental factors using geographically weighted regression. Environ.
several UHI studies, the indicators must be quantified in exactly Model. Software 25, 1789–1800.
the same way. Comparing different cities by means of different Lindberg, F., 2007. Modelling the urban climate using a local governmental geo-
database. Meteorol. Appl. 14, 263–273.
indicators for the UHI will likely lead to questionable results. As a Lowry, W.P., 1977. Empirical estimation of urban effects on climate – problem anal-
consequence, such studies should state the methods for data acqui- ysis. JApMe 16, 129–135.
sition and account for environmental and experimental conditions, Luo, Z.K., Sun, O.J., Ge, Q.S., Xu, W.T., Zheng, J.Y., 2007. Phenological responses of
plants to climate change in an urban environment. Ecol. Res. 22, 507–514.
which Stewart (2011) has already described; they should also make
Mostovoy, G.V., King, R.L., Reddy, K.R., Kakani, V.G., Filippova, M.G., 2006. Sta-
transparent the exact quantification of the UHI. If possible, several tistical estimation of daily maximum and minimum air temperatures from
indicators for the UHI should be presented in parallel. MODIS LST data over the state of Mississippi. GISci. Remote Sens. 43,
78–110.
Oke, T.R., 1982. The energetic basis of the urban heat-island. QJRMS 108, 1–24.
Acknowledgements Prihodko, L., Goward, S.N., 1997. Estimation of air temperature from remotely sensed
surface observations. Remote Sens. Environ. 60, 335–346.
We acknowledge the thermal maps, which were acquired Rajasekar, U., Weng, Q.H., 2009. Urban heat island monitoring and analysis using
a non-parametric model: a case study of Indianapolis. ISPRS J. Photogramm.
and processed by the engineering consultants Steinicke and Remote Sens. 64, 86–96.
Streifeneder, Freiburg (Germany). We thank Michael Naumann and Roetzer, T., Wittenzeller, M., Haeckel, H., Nekovar, J., 2000. Phenology in central
the Department of Environment, Section Urban Ecology, of the city Europe – differences and trends of spring phenophases in urban and rural areas.
IJBm 44, 60–66.
administration of Leipzig for approving the scientific usage of this Runnalls, K.E., Oke, T.R., 2000. Dynamics and controls of the near-surface heat island
data set. We thank the Saxon State Office for Environment, Agricul- of Vancouver, British Columbia. PhGeo 21, 283–304.
ture and Geology and the German Weather Service for providing Schwarz, N., Lautenbach, S., Seppelt, R., 2011. Exploring indicators for quantifying
surface urban heat islands of European cities with MODIS land surface temper-
their meteorological data and the US Geological Survey for pro- atures. Remote Sens. Environ. 115, 3175–3186.
viding the elevation data. The research was supported by TERENO Shustack, D.P., Rodewald, A.D., Waite, T.A., 2009. Springtime in the city: exotic shrubs
(Terrestrial Environmental Observatories). Furthermore, we thank promote earlier greenup in urban forests. Biol. Invasions 11, 1357–1371.
Stafoggia, M., Schwartz, J., Forastiere, F., Perucci, C.A., The SISTI Group, 2008. Does
Andreas Foth, Sebastian Fritsche, Michael Krüger, Sebastian Mayer, temperature modify the association between air pollution and mortality? A
Kerstin Stark and Jens Weinert for acquiring the field data. Finally, multicity case-crossover analysis in Italy. Am. J. Epidemiol. 167, 1476–1485.
we thank all the families who allowed us to mount weather stations Stewart, I.D., Oke, T., 2009. A new classification system for urban climate sites. Bull.
Am. Meteorol. Soc. 90, 922–923.
on their properties and two anonymous referees for very helpful
Stewart, I.D., 2011. A systematic review and scientific critique of methodology in
comments on an earlier version of this manuscript. modern urban heat island literature. IJCli 31, 200–217.
Streutker, D.R., 2002. A remote sensing study of the urban heat island of Houston,
Texas. Int. J. Remote Sens. 23, 2595–2608.
References Tomlinson, C.J., Chapman, L., Thornes, J.E., Baker, C.J., 2010. Derivation of Birming-
ham’s summer surface urban heat island from MODIS satellite images. IJCli,
Aniello, C., Morgan, K., Busbey, A., Newland, L., 1995. Mapping micro-urban heat doi:10.1002/joc.2261(online).
islands using Landsat TM and A GIS. Comput. Geosci. 21, 965–969. United States Geological Survey. GTOPO 30.
Arnfield, A.J., 2003. Two decades of urban climate research: a review of turbulence, http://eros.usgs.gov/#/Find Data/Products and Data Available/gtopo30 info.
exchanges of energy and water, and the urban heat island. Int. J. Climatol. 23, Voogt, J.A., Oke, T.R., 2003. Thermal remote sensing of urban climates. Remote Sens.
1–26. Environ. 86, 370–384.
Chen, X.L., Zhao, H.M., Li, P.X., Yin, Z.Y., 2006. Remote sensing image-based analysis of Weng, Q., Liu, H., Lu, D., 2007. Assessing the effects of land use and land cover patterns
the relationship between urban heat island and land use/cover changes. Remote on thermal conditions using landscape metrics in city of Indianapolis, United
Sens. Environ. 104, 133–146. States. Urban Ecosyst. 10, 203–219.
Cheng, K.S., Su, Y.F., Kuo, F.T., Hung, W.C., Chiang, J.L., 2008. Assessing the effect of Zhang, J.Q., Wang, Y.P., 2008. Study of the relationships between the spatial extent
landcover changes on air temperature using remote sensing images – a pilot of surface urban heat islands and urban characteristic factors based on Landsat
study in northern Taiwan. Landscape Urban Plan. 86, 85–96. ETM plus data. Sensors 8, 7453–7468.

You might also like